

180 episodes
by Latent.Space · English
The main Latent Space podcast: AI engineering, Forward Deployed Engineering, AI for science, and ⚡ Lightning episodes. Other feeds: Engineering and FDE: https://podcast.latent.space/eng.xml; Science: https://podcast.latent.space/science.xml; Misc (Lightning and other shows): https://podcast.latent.space/misc.xml; Firehose (all available shows): https://podcast.latent.space/firehose.xml. Feed directory: https://latent.space/pods

Alex Zhang (MIT) discusses GPU-kernel optimization communities (GPU mode), how AI is changing kernel development and verification, and why “harnesses” and recursively calling language models (RLMs)…
2 Oct 2026 · 1h 41m · 36 chapters
The episode is a Dev Day discussion about OpenAI’s computer use agents and new platform APIs, plus a debate about whether “computer use hasn’t advanced in two years.” Ari (product/engineering lead…
30 Sep 2026 · 39m · 17 chapters
Anthropic’s Claude Code “next era” with agentic coding, harness evolution, and new interfaces (Ask User Question, artifacts, CloudTag/projects, and CloudMods).
29 Sep 2026 · 1h 32m · 38 chapters
OpenRouter’s “PubSub for inference” model—how it turns swapping LLMs into a product marketplace with continuous routing, subscriptions to model “SKUs,” and an “auto router” that can switch models…
25 Sep 2026 · 1h 21m · 39 chapters
Runway’s evolution from using open-source generative models as creative tools to building real-time “world models” that can simulate dynamics and power interactive software.
25 Sep 2026 · 1h 36m · 41 chapters
The episode argues that AI-enabled DNA design is creating a “bio-security arms race,” and that defenses must advance alongside design.
23 Sep 2026 · 1h 32m · 42 chapters
AI for science via “scorable tasks” in Google’s Latents-Based Science/ERA.
22 Sep 2026 · 2h 1m · 52 chapters
Diogo Almeida (TypeSafe AI) launches Jev, describing “System 1” machine-native, large-programmable models meant for code consumption (“type safe”), optimized for intelligence per dollar.
21 Sep 2026 · 2h 21m · 61 chapters
Liquid AI’s worm-inspired continuous-time “liquid neural networks” and how Liquid Foundation Models (LFM) are architecturally searched and scaled for efficient, production-grade on-device and…
18 Sep 2026 · 1h 10m · 28 chapters
AIUC’s “confidence infrastructure” for frontier AI/agents—arguing that adoption is blocked less by capability and more by liability, risk, and trust.
16 Sep 2026 · 1h 26m · 40 chapters
Richard Socher (Recursive) discusses “the Eureka Machine,” a goal-directed superintelligence meant to invent most things for humanity, plus how to pace AI progress, regulate safely, and build…
14 Sep 2026 · 1h 32m · 42 chapters
AI for science—how to model the physical world with neural operators (especially Fourier Neural Operators) and how to verify neural networks using Lean/TorchLean.
26 Aug 2026 · 1h 24m · 29 chapters
State of the art for voice AI/voice agents in 2026, focusing on “forward deployed engineering” lessons: why most systems use cascaded pipelines (STT → LLM → TTS) instead of end-to-end…
25 Aug 2026 · 37m · 19 chapters
Simulation as a “new scaling law” for modeling human behavior, not just predicting outcomes.
21 Aug 2026 · 1h 10m · 25 chapters
EXO’s “fully recursive” self-improving agent architecture, arguing RSI is now feasible because agents iterate on their own harness code (not model weights).
15 Aug 2026 · 47m · 21 chapters
Chai Discovery’s “BioAI phase shift” from slow, trial-and-error protein binder discovery to a more loop-like, design-and-validate workflow enabled by Chai’s structure prediction and protein design…
11 Aug 2026 · 1h 35m · 39 chapters
How Baseten’s “inference engineering” speeds up LLM serving, especially for very long prompts, tool calling/structured outputs, and production readiness across new model releases (quantization,…
3 Aug 2026 · 1h 41m · 42 chapters
Akshay Nathan (OpenAI) explains how ChatGPT Work merges Codex and ChatGPT into a single “super app” for knowledge work, driven by model capability, shared agent “harness,” and product decisions that…
28 Jul 2026 · 1h 9m · 31 chapters
Eiso Kant (Poolside AI) explains how Poolside builds open foundation models using a “model factory” approach—industrializing the full pipeline from streaming pretraining data to post-training—so…
23 Jul 2026 · 1h 55m · 46 chapters
Zara Therapeutics’ XCell “virtual cell” model for drug discovery, arguing that causal prediction requires causal perturbation data (not just descriptive single-cell data).
21 Jul 2026 · 1h 30m · 36 chapters
Lila Sciences’ “AI science factories” for post-training scientific reasoning at scale.
16 Jul 2026 · 1h 41m · 48 chapters
Origin and strategy of the AI engineering conference (AIE) and “frontier lab” dynamics; plus discussion of inference chips (Etched), Anthropic’s Fable 5, OpenAI’s reported 5% equity stake to the US…
10 Jul 2026 · 28m · 12 chapters
Modal’s CTO argues that AI infrastructure must evolve from developer experience to “agent experience” by using self-provisioning runtimes that collocate infrastructure requirements with code,…
8 Jul 2026 · 58m · 25 chapters
How Genesis Molecular AI uses diffusion-based 3D structure prediction (Perl) plus physics-guided “inference-time thinking” to improve protein–small molecule interaction modeling for drug discovery,…
1 Jul 2026 · 1h 49m · 42 chapters